---
title: "Discussion on \"AWS ECS auto-scaler is broken\n(don’t worry, we’ve fixed it) \" | Hashnode"
description: "Discussion on \"AWS ECS auto-scaler is broken\n(don’t worry, we’ve fixed it) \". How Platformatic ICC Outperforms AWS ECS Target Tracking and Step Scaling"
url: "https://hashnode.com/posts/aws-ecs-autoscaler-vs-platformatic-icc/6a21a6ed47270a41c4f62bf8"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [AWS ECS auto-scaler is broken (don’t worry, we’ve fixed it)](https://blog.platformatic.dev/aws-ecs-autoscaler-vs-platformatic-icc?utm_source=hashnode&utm_medium=feed)

How Platformatic ICC Outperforms AWS ECS Target Tracking and Step Scaling

Jun 4 · 25 min read · blog.platformatic.dev  
Updated Jul 13  
Read the article

-   [#aws](https://hashnode.com/tag/aws)
-   [#ecs](https://hashnode.com/tag/ecs)
-   [#nodejs](https://hashnode.com/tag/nodejs)
-   [#platformatic](https://hashnode.com/tag/platformatic)
-   [#cloud-computing](https://hashnode.com/tag/cloud-computing)
-   [#performance](https://hashnode.com/tag/performance)
-   [#devops](https://hashnode.com/tag/devops)
-   [#infrastructure](https://hashnode.com/tag/infrastructure)

## Written by

[Ivan Tymoshenko](https://hashnode.com/@ivan-tymoshenko)

@ivan-tymoshenko · Warsaw, Poland

  
[View profile](https://hashnode.com/@ivan-tymoshenko)

## Responses 1

Markdown supported · ⌘↵ to post

[Atul Vishwakarma](https://hashnode.com/@vatul16)

[Jun 4](https://hashnode.com/posts/aws-ecs-autoscaler-vs-platformatic-icc/6a21a6ed47270a41c4f62bf8/comment/6a21ae1f09761aac249266ba)

This is a really insightful comparison! The distinction you draw between AWS ECS's traditional autoscaling (which reacts to metrics) and Platformatic ICC's predictive/intelligent approach is super clear.

What stood out to me the most:

The latency argument: Waiting for CloudWatch metrics + ECS to react means your app already feels slow to users. ICC's proactive scaling actually makes sense for real-world user experience.

Cost transparency: Breaking down the hidden costs of over-provisioning vs. under-provisioning is something most comparisons skip.

I'm curious though — for smaller startups or side projects where traffic is unpredictable but not necessarily high-volume, would ICC still be cost-effective? Or does the sweet spot come at a certain scale threshold?

Also, the platformatic upgrade command being that simple is tempting 😄 Has anyone tried migrating an existing ECS service to ICC in production? Would love to hear war stories if so.

Great write-up, definitely making me rethink my infrastructure choices!